腾讯混元发布 WeVisOne-VL-4B 统一视觉语言模型

WeVisOne-VL: Unified Visual Perception from Objects to Pixels

WeVisOne-VL (WeChat Vision All-in-One) is a unified vision-language model developed by WeChat Vision, Tencent Inc. Built upon Qwen3-VL, it unifies object-level perception (open-vocabulary object detection, referring localization, object counting, object pointing, and OCR) and pixel-level perception (referring segmentation and promptable alpha matting) through structured next-token prediction and a lightweight 9.6M-parameter pixel decoder.

Quick Start

Install the inference package from GitHub, then:

from PIL import Image
from wevisone_vl import WeVisOneVL

model = WeVisOneVL("tencent/WeVisOne-VL-4B", device="auto", dtype="bfloat16")
image = Image.open("example.jpg").convert("RGB")
result = model.predict(image, task="matting", query="person", temperature=0.0)
print(result.response, len(result.masks))

Architecture & Released Revision

The model uses dedicated coordinate tokens (<0> ... <1000>) to represent normalized bounding boxes and points, returns None when a queried target is absent, and emits a special instance token (<OBJ>, token ID 152678) after each predicted box for dense tasks.

The attached mask_decoder.pt (9,609,764 parameters) is a required core component that combines:

A Visual Feature Adapter over multi-level Qwen3-VL pre-merging ViT features;
A Box-Guided Query Decoder with 6 deformable cross-attention blocks initialized from the <OBJ> hidden state and Fourier-encoded box prior;
An Instance-Conditioned Dense Prediction Head with FullCanvasAlphaCascadeHead (refining predictions across S/4, S/2, and S/1 to output full-resolution continuous alpha mattes at stride 1).

Intended Use

The release is intended for research in visual grounding, referring localization, pointing, counting, OCR localization, referring segmentation, and image matting. It is not intended to make autonomous high-impact decisions or to identify individuals.

Limitations

Fine alpha boundaries remain sensitive to image resolution, severe occlusion, complex transparency, and prompt phrasing.
For inference efficiency, the model directly generates structured outputs without explicit chain-of-thought (CoT) reasoning, and currently targets single-turn image perception (video grids are rejected by the pixel decoder).

License

Unless otherwise noted, Tencent-authored code in this repository is released under the Apache License 2.0.

Third-party components retain their original licenses and copyright notices. Please review the corresponding source files before use.